Dell Enterprise Hub at Dell Tech World 2026: new models, new platforms, faster to production
Last week at Dell Technologies World, Michael Dell announced the next wave of models in Dell Enterprise Hub (DEH), Dell and Hugging Face's on-premises home for open-source AI, ready to deploy on Dell infrastructure from day one. Two stories run through this release.
The open-source frontier keeps moving, DEH moves with it, on every Dell platform. Roughly twenty new model configurations have landed in recent months (DeepSeek V4, Kimi K2.6, GLM 5.1, MiniMax M2.7, the Nemotron 3 reasoning family, Gemma 4, Qwen 3.5 and more) optimized for Dell AI Servers and AI PCs, with tested, secured containers ready within hours of each public release.
Moving from model selection to production is easier than ever with DEH. This release narrows the gap with three things: benchmarked deployment configurations for each Dell platform (Goodput Scenarios, now extended to NVIDIA B300 and GB10), signed and scanned container images you can trace and verify, and new dell-ai SDK utilities that inspect a platform's configuration before a single container is pulled.
Try it today at dell.huggingface.co.
Latest open-source models, day one, on every Dell platform
Dell Enterprise Hub is a multi-platform model catalog moving with the open-source frontier. It offers ready-to-use, optimized deployments of the latest models across the full range of enterprise AI hardware, from datacenter to workstation. Here is the concrete picture for Dell Tech World: what is new in the catalog, which Dell platform configurations it runs on, and where it ships with a tuned Goodput Scenario (marked ★).
| Model | Dell PowerEdge Platform configurations |
|---|---|
deepseek-ai/DeepSeek-V4-Pro |
XE9680-H200 ★, XE9785-B300 ★ |
deepseek-ai/DeepSeek-V4-Flash |
XE9680-H100 ★, XE9680-H200 ★ |
zai-org/GLM-5.1 |
XE9785-B300 ★ |
zai-org/GLM-5.1-FP8 |
XE9680-H200, XE9680-MI300X, XE9785-MI355X |
moonshotai/Kimi-K2.6 |
XE9680-H200, XE9680-MI300X, XE9785-MI355X, XE9785-B300 ★ |
nvidia/Kimi-K2.6-NVFP4 |
XE9680-H200, XE9785-B300 |
MiniMaxAI/MiniMax-M2.7 |
XE9680-H100, XE9680-H200, MI300X, XE9785-MI355X |
nvidia/MiniMax-M2.7-NVFP4 |
XE9680-H100 ★, XE9680-H200 ★, XE9785-B300 |
nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-FP8 |
Dell Pro Max with GB10 ★ |
google/gemma-4-31B-it |
Dell Pro Max with GB10 |
Qwen/Qwen3.5-27B |
Dell Pro Max with GB10 |
Alongside the headline MoEs, the catalog also picked up Arcee Trinity Large Thinking (a ~398B sparse MoE with native long chain-of-thought), the rest of the Google Gemma 4 family (including the new MTP speculative decoding enabled versions), Mistral Small 4 119B (and its NVFP4 sibling), Qwen3 Coder Next, and Cohere Transcribe 03-2026 for high-quality on-prem ASR. GLM 5.1 NVFP4 is coming soon.
Goodput Scenarios: from "deployed" to "deployed at SLO"
A model running is not the same as a model running well. So what is Goodput? The number of served requests per second a deployment can sustain while still meeting the service level objectives (SLO) for a real workload (chat, RAG, agentic, long-context) not just the peak number on a benchmark chart. We made the broader case for the metric earlier this year in From Benchmark Theater to Real Performance: A Case for Goodput.
Goodput Scenarios are how DEH operationalizes that. Instead of spending a week iterating on vLLM flags, you start from a configuration Dell and Hugging Face have benchmarked together to hit the target SLO on the platform of your choice. Each scenario fixes the SLO (max model context, virtual users, input and output token ranges) and exposes the vLLM parameters benchmarked to meet it on that platform.
On H100, H200, B300 and L40s, you pick one of three named scenarios:
- Balanced: typical chat and content workloads, moderate context and concurrency.
- High concurrency: high request volume, smaller context, optimized for throughput.
- Long context: long-document, agentic, or RAG-heavy workloads with extended context windows.
New for Dell Tech World, Goodput Scenarios are now published for NVIDIA B300 and GB10, joining existing coverage on H100, H200 and L40s. B300 ships with the full three scenarios; GB10 ships with a single "Performance" scenario today.
The full methodology, SLO tables and per-platform configurations are in the Goodput Scenarios guide.
The dell-ai SDK: deployment-day verification, not just an API client
Dell Enterprise Hub is a portal, but enterprises need to script, automate and verify. That is what the dell-ai SDK is for. It is an open-source Python library and CLI that Dell and Hugging Face publish together, so customers can drive DEH (browse the catalog, generate deployment snippets, inspect platform configs) from a terminal or a CI pipeline rather than only from the web UI.
Two capabilities have taken dell-ai from a thin API client into a production tool this year.
Infrastructure verification. dell-ai utils describe-system dumps the current host as structured JSON: kernel, CPU, GPU model and driver, CUDA toolkit, NVIDIA Container Toolkit, Kubernetes server, storage layout. dell-ai utils check-system then compares your host against the canonical system profile DEH publishes for each Dell platform SKU. The two are co-designed: when DEH ships support for a new platform, the SDK can verify a customer's actual machine matches the reference before a single container is pulled.
Faster, scriptable model discovery. dell-ai models search now runs filters server-side, fans out detail fetches in parallel, and caches model metadata on disk for 24 hours. A cold search across the full catalog drops from ~16 seconds to ~2, and warm searches return in ~100 ms. New commands like dell-ai models compatible-platforms <model_id> and a --format table flag make the CLI results easy readable.
With the CLI, a day-one deployment experience would look like this:
dell-ai utils check-system # verify infra matches DEH reference
dell-ai models search --platform-id xe9785-amd-mi355x # what runs on this hardware?
dell-ai models compatible-platforms moonshotai/Kimi-K2.6 # where can we run this model?
# How can I run this model?
dell-ai models get-snippet -m deepseek-ai/DeepSeek-V4-Pro -p xe9785-nvidia-b300 --gpus 4
Together, we're just getting started
With the new Dell PowerEdge XE9785 (AMD MI355X and NVIDIA B300) and Dell Pro Max with GB10, a refreshed catalog of frontier open-source models on day zero, Goodput Scenarios that translate raw throughput into SLO-aware deployments, and an SDK that verifies your infrastructure before you deploy, Dell Enterprise Hub is the fastest path from "model released yesterday" to "model running in our private datacenter or workstation today".
Try it today at dell.huggingface.co.
